In an AI comparison table, a fact a location never stated is an empty cell.

Google's Hydrogen upgrade brings more comparison tables to Search. For a business with multiple locations, the bigger risk is not fewer clicks. It is a blank row sitting next to a sibling branch that stated its hours.

· 10 minute read

Google said on November 12 that Gemini 4 Hydrogen now runs Search for all users worldwide, and that the change brings more structured responses and comparison tables. Most coverage will treat this as news about clicks and citations. For a business with multiple locations, the more useful way to read it is in terms of columns. A comparison table gives every business the same columns. A fact that a location never published does not fade into a paragraph. It shows up as a blank, in a row right next to a competitor that filled it in, or next to your own branch across town that did.

Google's Hydrogen upgrade brings comparison tables to every Search user

Google's announcement of Gemini 4 Hydrogen in Search, written by Robby Stein, VP of Product for Google Search, makes two claims that matter to anyone comparing businesses. The first is that the upgrade strengthens query fan-out, where Search breaks a question into related subtopics and runs queries across all of them at the same time. The second is that users will see more images, structured responses and comparison tables.

Google did not publish how a comparison table is built, which sources fill which cells, or what happens when a value is missing. This essay does not claim to know any of that. What can be said with confidence is narrower but still useful: by its form, a table asks the same questions of every row. Whatever the model fills in, it fills in against shared columns.

That form is the part of the Hydrogen upgrade that a business with branches, franchise units or regional offices can actually act on. The columns are the same for everyone. What each location supplies to fill them is up to the business.

A comparison table reads fields, not prose

A comparison table reduces each business to a row of fields: when it is open, where it works, what qualifies it, and how it charges. Prose can carry those facts, but prose can also avoid them gracefully. A location page that says “we proudly serve the greater metro area with flexible scheduling” reads fine as a paragraph. Put that page in a table with a Service area column and an Hours column, and it gives you nothing to put in either cell.

The difference is about extraction, not about the quality of the writing. A fact stated outright, such as “Open Monday to Saturday, 7 a.m. to 6 p.m.”, can go straight into a cell without interpretation. A fact only implied, such as “flexible scheduling”, forces the answer engine to guess, look somewhere else, or leave the cell empty. All three outcomes are worse for the location than stating the fact, and the last one is visible to the reader in a way a vague sentence never was.

That is why adding content does not fix a weak comparison. A thousand more words about service quality do not fill an Hours column. One line does.

The empty cell often sits next to your own branch

For a business with multiple locations, the most damaging comparison is often an internal one. Someone asking for the nearest provider of a service may be shown two or three of the same company's branches alongside one or two peers, and the table treats each branch as its own row. If one location page states its hours, service area, license and pricing basis and the next one states none of them, the table shows the same brand looking complete in one row and empty in the next.

Here is an illustration, made up for this essay rather than taken from a measured client. A heating and cooling company runs three branches. The first branch page lists its hours, the counties it covers, a state contractor license number, and the line “flat diagnostic fee, then a written quote.” The second lists hours and counties but no license. The third was built from an older template and says only that the team is “ready to help, call today.” In a side by side table, the first branch reads as a full row and the second as a row with one gap. The third is a row of blanks, and a peer down the road with a plain, complete page will look stronger in every column.

Nothing about the third branch is actually worse. It may be the busiest of the three. The table has no way to know that, because it compares what was stated, and the third branch stated nothing that could be compared.

Four comparable facts carry most of a local comparison

The comparable facts for a local business are the ones a person would use as column headers when choosing between providers: hours, service area, credentials and pricing basis. Each one is a field rather than a story, and each one already has an established structured form.

Hours are the clearest case. Google's documentation for local business structured data says to define each location as its own LocalBusiness, each with its own properties such as opening hours or telephone numbers, and it handles departments inside a store the same way, defining separately any property that differs from the main store. Service area has a defined form too: schema.org describes areaServed as the geographic area where a service or offered item is provided. Credentials mean a license number, a certifying body or an accreditation, stated as a name and a number rather than as “fully qualified.” Pricing basis means the method, not the price: flat fee, hourly, free estimate, or quote after inspection.

Structured data is one way to state these facts, and the visible page is another. The markup itself is not the point. The point is that each fact exists on the page for that location, in words a machine can lift without having to infer anything.

Stating facts identically at every location means the same fields, not the same values

For a multi-location business, identical means every location page answers the same four questions in the same form, in the same place, under the same labels. The values will differ. The downtown branch may close at 8 and the suburban one at 5. One may cover two counties and another six. A comparison handles different values well. It handles missing values badly, and it handles inconsistent formats unpredictably: “weekdays till 5ish” on one page and a structured hours block on another give an answer engine two different extraction problems for what should be the same field.

The practical test is a store locator audit. Open every location page in the locator and write down the four facts for each one. If you cannot write a fact down from the page alone, that location will read as incomplete in a table. If the fact is there but worded differently from the sibling pages, that location depends on the model interpreting it correctly. A template that makes every location page state the same four fields closes both gaps at once, without adding a single paragraph.

This is also where a regional or franchise brand has an advantage over an independent peer, if it uses it. One template applied to forty locations produces forty complete rows. Without that template, the same scale produces forty different ways to leave a cell empty.

The strongest objection: models read prose well and tables draw on more than your site

The strongest case against this thesis has two parts. First, modern language models read prose very well, and a model that can reason across subtopics can infer hours from “open early, closed Sundays” about as easily as from a formatted block. Second, a comparison table in Search is unlikely to rely on a location page alone. Business listings, review sites and other third parties also carry hours and service areas, and a cell could be filled from any of them.

The first objection is partly right, but it still does not rescue the vague page. Inference works when the fact is present in some form. “Open early, closed Sundays” contains part of a fact. “Ready to help, call today” contains none, and no amount of reasoning recovers a fact that was never written down. Even when inference works, it produces an approximation, and an approximation in one row next to an exact value in the next row reads as the weaker answer.

The second objection is the more serious one. Honestly answered, it changes the stakes but not the conclusion. If third parties fill the cell, the location page is no longer the only source, but it is still the only source the business fully controls. A location that states its own facts gives every other source something to agree with. A location that states nothing leaves the cell to whichever listing was updated most recently. For a business with many branches, that is often a listing nobody at the branch has looked at in years.

A third, quieter objection is that templated location pages look thin. That risk is real when a template replaces substance with swapped city names. It does not apply to four factual fields that really do differ from one location to the next. A license number and a service area are the opposite of filler.

What an on-page comparison can show, and what it cannot

An on-page comparison can show whether each location states each comparable fact, and how its pages stack up against a sibling branch or a public peer on the same fields. That is the comparison AIOStory is built for. AIOTruth evaluates each site deterministically, and AIOStory compares the results across a business's locations and against peers, showing where a brand's story holds, where it changes, and where it loses ground.

An on-page comparison cannot show what a Hydrogen comparison table actually displayed to a particular user on a particular day. AIOStory does not capture live model answers, and a deterministic scan of a location page is never presented as one. The link between a missing field on a page and an empty cell in an answer is an inference from how tables work, not a measured rate. There is not yet any evidence about how often Hydrogen tables show a blank where a field is missing, and this essay does not claim to have any.

What a deterministic comparison does offer is a result someone else can reproduce: the same location pages, checked the same way on the same day, produce the same table of fields. Coverage is reported as N of M locations read, and a location page that cannot be read at all is reported as a finding in its own right. An unreadable location page is the emptiest row there is.

A thousand more words about service quality do not fill an Hours column. One line does.

Most people will read the Hydrogen upgrade as a story about clicks. For a business with branches, it is a story about rows. Every location can now be compared field by field with the one next to it, whether that neighbor is a competitor or a sibling branch under the same name. The fix is small, unglamorous and easy to check. Decide the four facts every location must state. State them the same way on every location page. Then check each one against the page itself, not against what the team assumes is there.

Further reading

  1. Google, Robby Stein's announcement that Gemini 4 Hydrogen runs Google Search for all users, describing stronger query fan-out and more comparison tables blog.google/products-and-platforms/products/search/gemini-4-hydrogen-search-upgrade
  2. Google Search Central, local business structured data documentation, the primary guidance on defining each location with its own hours and properties (no publish date on page; accessed date shown) developers.google.com/search/docs/appearance/structured-data/local-business
  3. Schema.org, definition of the areaServed property used to state a service area (date is the schema version stamp) schema.org/areaServed

Every source above was fetched and a verbatim phrase confirmed on the page before this essay published. Nothing here is paraphrased from memory.